It depends on what you want to get.
If it is per interacting alpha, you have to divide your event-by-event counter by 922 (i.e., the total number of simulated inelastic interactions).
If it is per incident alpha, you have to divide it by 1E6 and multiply it by 0.02.
My concept regarding the use of number of stars is not very clear. I got some confusion.
Total number of inelastic interactions = Int
Total weight of the inelastic interactions = W
History = H
no of stars per beam particle = W/H
For a nuclear interaction, if I want to calculate neutron dose at a particular location using USRBIN or want to score neutron flux by USRBDX, the result is in per primary, in such cases, we do not need to multiply by no of stars.
However, while using USRYIELD card, to plot double differential sigma, we are multiplying the result by no of stars, i.e. W/H
What is the reason for this? Also, if we score double differential N instead of sigma, do we also have to multiply the result by no of stars (W/H)?
Now say, LAM Bias is used. We are using USRYIELD card. To plot double differential sigma, we are still multiplying the result by no of stars, i.e. W/H.
But when mgdraw.f is used to count no of neutrons, we are multiplying the result by W/(H*Int)
This is because here you want to (exceptionally) obtain a microscopic cross section. Thereby, based on the knowledge of the reaction cross section (which is the embedded multiplication factor, linked to the inelastic interaction probability), you need to get a yield per reaction (star, and not per incident particle), to be multiplied by the reaction cross section.
By the way, note that you need to divide (and not multiply) by the number of stars per beam particle, namely the star probability.
Here, again, it depends on what you are interested in. For normal applications, you may be interested in the yield per incident particle, as naturally provided by USRYIELD. But if you aim to transform this into a microscopic cross section (when it makes sense, i.e., for a thin target), then we are back to the above.
The reason is pretty obvious (filtering the reiterated multiplication vs division mistake): USRYIELD takes automatically into account the biasing factor, which mgdraw has to leave on user coding responsibility.